skills/daxaur/openpaw/c-research/Gen Agent Trust Hub

c-research

Pass

Audited by Gen Agent Trust Hub on Mar 9, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted data from various external sources, making it vulnerable to indirect prompt injection. Malicious instructions within summarized content could be used to manipulate the agent's behavior.
  • Ingestion points: Web URLs, PDF files, YouTube videos, and podcast feeds referenced in SKILL.md.
  • Boundary markers: No delimiters or instructions are used to separate user data from system instructions.
  • Capability inventory: Execution of the 'summarize' CLI tool (SKILL.md).
  • Sanitization: There is no evidence of data sanitization or filtering.
  • [COMMAND_EXECUTION]: The skill directs the agent to execute a CLI tool with absolute file paths (SKILL.md), which could be exploited to access and expose sensitive local system files if the agent has sufficient permissions.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 9, 2026, 08:44 AM
Security Audit — agent-trust-hub — c-research